thanks! as i have written in the other thread, i am evaluating the application of PyMC3 for EEG and MEG analysis. Doing it like this would already be a huge leap forward for the community.
However, as you mentioned the covariance matrix being diagonal: This is clearly not the case for our sensors. In fact they are highly correlated. Currently this is, if at all, taken into account indirectly.
So: If I know the noise-covariance matrix of my channels, how would I need to change the above model in order to reflect this?
I am aware that in this case, I would not be able to split the computation anymore…